There is a number that has been haunting my sleep lately: $281 billion. It's not a market cap, not a GDP figure, and not some nation's stimulus package. It's Goldman Sachs' projected global wafer fab equipment (WFE) spending for 2028. And the more I stare at it, the more I realize the number itself isn't the story. The silence around it is.
Finding the signal in the silence of the bear taught me that the most important data points are often the ones nobody is talking about. In this case, it's not the staggering 37% CAGR from 2026 to 2028 that whispers secrets. It's the quiet assumptions buried beneath the forecast—assumptions about export controls, about AI's staying power, and about the very nature of a supply chain that has become the world's most expensive geopolitical chessboard.
This isn't a report about machines. It's a story about belief, about the narratives we construct around silicon, and about what happens when those narratives collide with reality.
The Context: A Supercycle's Skeleton
To understand why Goldman's forecast matters, you have to understand the machinery of modern chipmaking. A single leading-edge fab costs upwards of $20 billion. The equipment inside it—the EUV lithography scanners, the etch tools, the deposition chambers—accounts for 70-80% of that capital expenditure. These are not off-the-shelf purchases. A single EUV machine from ASML costs around $180 million and takes 12-18 months to deliver. The lead time for the next-generation High-NA EUV systems stretches to two years.
This is the world of WFE, a market currently worth around $1,000-1,200 billion annually, dominated by a handful of oligopolies: ASML in lithography, Applied Materials and Lam Research in etch and deposition, KLA in metrology. Their customers are an even more exclusive club—TSMC, Samsung, Intel, SK Hynix, and Micron—who together account for over half of all equipment revenue.
The current cycle is being driven by an insatiable appetite for AI compute. Advanced logic nodes (5nm and below) are running at 95%+ utilization. HBM memory, the high-bandwidth stacks that feed AI accelerators, is in a supply crunch that won't ease before 2028. TSMC's CoWoS advanced packaging capacity—the bottleneck for NVIDIA's GPU shipments—is being doubled, then doubled again.
Goldman's forecast essentially says this party doesn't just continue; it accelerates. From $150 billion in 2026 to $218 billion in 2027, then to $281 billion in 2028. That's not a cycle. That's a supercycle.
The Core: Two Engines, One Narrative
Decoding the hidden stories behind the tokenomics of this forecast reveals something most analysts miss: the WFE market is no longer a single-engine aircraft. It's becoming a twin-engine jet.
The first engine is logic. TSMC's Arizona fabs, Samsung's Taylor complex, and Intel's global expansion are all pulling in leading-edge equipment. The second, and arguably more powerful, engine is memory—specifically HBM. And here's the insight that changes everything: HBM production requires a completely different equipment set than logic.
TSV etching, electroplating, wafer bonding, temporary debonding—these are processes that don't overlap with traditional logic manufacturing. The equipment for HBM is a parallel universe of tools, and it's scaling at a pace that rivals the logic side. SK Hynix alone is investing $90 billion in a new semiconductor cluster in Yongin, South Korea, with four fabs planned by 2030. Micron is committing over $100 billion to new DRAM facilities in New York and Idaho.
This dual-engine dynamic means the WFE market is more resilient than in previous cycles. If logic spending dips, memory can carry the load, and vice versa. It's a diversification that didn't exist in the 2017-2018 memory boom or the 2020-2021 logic surge.
But there's a deeper layer to this narrative. The equipment itself is becoming a bottleneck. ASML's EUV output would need to jump from roughly 50 units in 2024 to 80-100 per year by 2028 to meet Goldman's implied demand. That's a massive assumption about ASML's ability to scale production of the most complex machine ever built—a machine with over 100,000 parts, many of which come from a single supplier (Zeiss for optics).
Based on my audit experience, I've seen how supply chain constraints in this industry don't just cause delays; they reshape the entire competitive landscape. When equipment is scarce, the fabs that get priority are the ones with the deepest pockets and the strongest relationships. This creates a feedback loop where the rich get richer, and the gap between the haves and have-nots widens.
The Contrarian Angle: The Geopolitical Elephant
Here's where the narrative gets uncomfortable. Goldman's $281 billion forecast contains a hidden assumption that I find deeply questionable: that export controls on China will remain "rational."
Let me put this in perspective. China currently accounts for roughly 30% of global WFE spending. If the US, Netherlands, and Japan continue to tighten restrictions—which they have done consistently since 2019—China's equipment purchases will shrink dramatically. The math simply doesn't work without China buying $400-500 billion worth of equipment annually by 2028.
But here's the contrarian twist: what if the export controls actually accelerate the WFE cycle? China's response to sanctions has been a massive push for domestic equipment self-sufficiency. The Big Fund III, with its 344 billion yuan ($48 billion), is pouring money into domestic toolmakers like Naura, AMEC, and ACM Research. These companies are currently at 20-25% domestic market share, targeting 50% by 2028.
If Chinese equipment makers achieve even modest breakthroughs in mature-node etch and deposition tools, they could actually increase China's total equipment spending—just from domestic suppliers instead of foreign ones. This would be a net positive for the global WFE market, just with a different distribution of winners.
The crash is just a chapter, not the end. The same applies to the geopolitical tensions that have defined this industry. The narrative of decoupling is real, but it's not a one-way street. It's creating new opportunities even as it destroys old ones.
The Hidden Risks: What the Forecast Ignores
Alchemy is just storytelling with better chemistry, and Goldman's forecast is a masterclass in narrative alchemy. But every story has its blind spots. Let me map the unspoken desires of the early adopters—and the risks they're ignoring.
First, there's the AI capex sustainability question. Goldman's forecast implies that cloud giants (Microsoft, Google, Amazon, Meta) will maintain their current AI spending trajectory through 2028. That's a combined $300+ billion in annual capex, with no signs of slowdown. But what happens if AI revenue doesn't materialize as quickly as expected? What if the ROI on AI infrastructure starts to look shaky by 2027?
The semiconductor industry has a long history of boom-bust cycles driven by over-optimism. The 2017-2018 memory boom ended in a brutal downturn. The 2020-2021 logic surge was followed by a 2022-2023 correction. The current AI-driven cycle is bigger than both, which means the eventual correction—if it comes—will be proportionally more painful.
Second, there's the equipment delivery bottleneck. Even if demand materializes, can the supply chain deliver? ASML's High-NA EUV program is behind schedule. KLA's metrology tools have 6-12 month lead times. The industry is already operating at maximum capacity, and scaling up takes years, not quarters.
Third, and this is the one that keeps me up at night: the depreciation cliff. New fabs coming online in 2027-2028 will carry massive depreciation loads. TSMC's Arizona fab, for example, will have initial gross margins below 40%—compared to the company's 55%+ average—due to depreciation and higher labor costs. This will pressure the entire industry's profitability, even as revenue grows.
The Takeaway: Listening to What the Data Refuses to Say
Where meme meets strategy, magic happens. And right now, the meme is "AI supercycle," and the strategy is "buy equipment stocks." But the magic—the real insight—is in what the data refuses to say.
Goldman's forecast is directionally correct but likely 10-15% too optimistic. The WFE market will grow substantially through 2028, driven by AI and HBM demand. But the growth will be lumpier, more contested, and more politically fraught than the smooth curve suggests.
The real opportunity isn't in the equipment makers themselves—they're already priced for perfection. It's in the second-order effects: the materials suppliers, the advanced packaging players, the domestic Chinese toolmakers who are quietly building capabilities that will matter in 2030.
Weaving viral moments into lasting lore is what this industry does best. The question is whether the lore of the AI supercycle will be a story of sustainable growth or another chapter in the semiconductor industry's long history of self-inflicted wounds.
The signal is there, if you listen to the silence. The question is whether anyone will hear it before the noise becomes deafening.
